2 * Copyright 2016 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
24 #ifndef SI_SHADER_PRIVATE_H
25 #define SI_SHADER_PRIVATE_H
27 #include "si_shader.h"
28 #include "gallivm/lp_bld_flow.h"
29 #include "gallivm/lp_bld_init.h"
30 #include "gallivm/lp_bld_tgsi.h"
31 #include "tgsi/tgsi_parse.h"
32 #include "ac_shader_abi.h"
33 #include "ac_llvm_util.h"
34 #include "ac_llvm_build.h"
36 #include <llvm-c/Core.h>
37 #include <llvm-c/TargetMachine.h>
39 struct pipe_debug_callback
;
40 struct ac_shader_binary
;
42 #define RADEON_LLVM_MAX_INPUT_SLOTS 32
43 #define RADEON_LLVM_MAX_INPUTS 32 * 4
44 #define RADEON_LLVM_MAX_OUTPUTS 32 * 4
46 #define RADEON_LLVM_MAX_SYSTEM_VALUES 11
47 #define RADEON_LLVM_MAX_ADDRS 16
49 struct si_shader_context
{
50 struct lp_build_tgsi_context bld_base
;
51 struct gallivm_state gallivm
;
52 struct ac_llvm_context ac
;
53 struct si_shader
*shader
;
54 struct si_screen
*screen
;
56 unsigned type
; /* PIPE_SHADER_* specifies the type of shader. */
58 /* For clamping the non-constant index in resource indexing: */
59 unsigned num_const_buffers
;
60 unsigned num_shader_buffers
;
62 unsigned num_samplers
;
64 /* Whether the prolog will be compiled separately. */
67 struct ac_shader_abi abi
;
69 /** This function is responsible for initilizing the inputs array and will be
70 * called once for each input declared in the TGSI shader.
72 void (*load_input
)(struct si_shader_context
*,
74 const struct tgsi_full_declaration
*decl
,
77 /** This array contains the input values for the shader. Typically these
78 * values will be in the form of a target intrinsic that will inform the
79 * backend how to load the actual inputs to the shader.
81 struct tgsi_full_declaration input_decls
[RADEON_LLVM_MAX_INPUT_SLOTS
];
82 LLVMValueRef inputs
[RADEON_LLVM_MAX_INPUTS
];
83 LLVMValueRef outputs
[RADEON_LLVM_MAX_OUTPUTS
][TGSI_NUM_CHANNELS
];
84 LLVMValueRef addrs
[RADEON_LLVM_MAX_ADDRS
][TGSI_NUM_CHANNELS
];
86 /** This pointer is used to contain the temporary values.
87 * The amount of temporary used in tgsi can't be bound to a max value and
88 * thus we must allocate this array at runtime.
92 LLVMValueRef system_values
[RADEON_LLVM_MAX_SYSTEM_VALUES
];
97 struct lp_build_if_state merged_wrap_if_state
;
99 struct tgsi_array_info
*temp_arrays
;
100 LLVMValueRef
*temp_array_allocas
;
102 LLVMValueRef undef_alloca
;
104 LLVMValueRef main_fn
;
105 LLVMTypeRef return_type
;
107 /* Parameter indices for LLVMGetParam. */
108 int param_rw_buffers
;
109 int param_const_and_shader_buffers
;
110 int param_samplers_and_images
;
111 int param_bindless_samplers_and_images
;
112 /* Common inputs for merged shaders. */
113 int param_merged_wave_info
;
114 int param_merged_scratch_offset
;
116 int param_vertex_buffers
;
117 int param_rel_auto_id
;
118 int param_vs_prim_id
;
119 int param_vertex_index0
;
120 /* VS states and layout of LS outputs / TCS inputs at the end
121 * [0] = clamp vertex color
123 * [8:20] = stride between patches in DW = num_inputs * num_vertices * 4
124 * max = 32*32*4 + 32*4
125 * [24:31] = stride between vertices in DW = num_inputs * 4
128 int param_vs_state_bits
;
129 int param_vs_blit_inputs
;
131 int param_streamout_config
;
132 int param_streamout_write_index
;
133 int param_streamout_offset
[4];
136 /* Layout of TCS outputs in the offchip buffer
138 * [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
140 * [6:11] = the number of output vertices per patch, max = 32
142 * [12:31] = the offset of per patch attributes in the buffer in bytes.
143 * max = NUM_PATCHES*32*32*16
145 int param_tcs_offchip_layout
;
148 /* Offsets where TCS outputs and TCS patch outputs live in LDS:
149 * [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
150 * [16:31] = TCS output patch0 offset for per-patch / 16
151 * max = (NUM_PATCHES + 1) * 32*32
153 int param_tcs_out_lds_offsets
;
154 /* Layout of TCS outputs / TES inputs:
155 * [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
156 * max = 32*32*4 + 32*4
157 * [13:18] = gl_PatchVerticesIn, max = 32
158 * [19:31] = high 13 bits of the 32-bit address of tessellation ring buffers
160 int param_tcs_out_lds_layout
;
161 int param_tcs_offchip_offset
;
162 int param_tcs_factor_offset
;
165 int param_tes_offchip_addr
;
168 int param_tes_rel_patch_id
;
170 int param_es2gs_offset
;
172 int param_gs2vs_offset
;
173 int param_gs_wave_id
; /* GFX6 */
174 LLVMValueRef gs_vtx_offset
[6]; /* in dwords (GFX6) */
175 int param_gs_vtx01_offset
; /* in dwords (GFX9) */
176 int param_gs_vtx23_offset
; /* in dwords (GFX9) */
177 int param_gs_vtx45_offset
; /* in dwords (GFX9) */
179 int param_block_size
;
181 LLVMTargetMachineRef tm
;
183 /* Preloaded descriptors. */
184 LLVMValueRef esgs_ring
;
185 LLVMValueRef gsvs_ring
[4];
186 LLVMValueRef tess_offchip_ring
;
188 LLVMValueRef invoc0_tess_factors
[6]; /* outer[4], inner[2] */
189 LLVMValueRef gs_next_vertex
[4];
190 LLVMValueRef postponed_kill
;
191 LLVMValueRef return_value
;
209 static inline struct si_shader_context
*
210 si_shader_context(struct lp_build_tgsi_context
*bld_base
)
212 return (struct si_shader_context
*)bld_base
;
215 static inline struct si_shader_context
*
216 si_shader_context_from_abi(struct ac_shader_abi
*abi
)
218 struct si_shader_context
*ctx
= NULL
;
219 return container_of(abi
, ctx
, abi
);
222 unsigned si_llvm_compile(LLVMModuleRef M
, struct ac_shader_binary
*binary
,
223 LLVMTargetMachineRef tm
,
224 struct pipe_debug_callback
*debug
);
226 LLVMTypeRef
tgsi2llvmtype(struct lp_build_tgsi_context
*bld_base
,
227 enum tgsi_opcode_type type
);
229 LLVMValueRef
bitcast(struct lp_build_tgsi_context
*bld_base
,
230 enum tgsi_opcode_type type
, LLVMValueRef value
);
232 LLVMValueRef
si_llvm_bound_index(struct si_shader_context
*ctx
,
236 void si_llvm_context_init(struct si_shader_context
*ctx
,
237 struct si_screen
*sscreen
,
238 LLVMTargetMachineRef tm
);
239 void si_llvm_context_set_tgsi(struct si_shader_context
*ctx
,
240 struct si_shader
*shader
);
242 void si_llvm_create_func(struct si_shader_context
*ctx
,
244 LLVMTypeRef
*return_types
, unsigned num_return_elems
,
245 LLVMTypeRef
*ParamTypes
, unsigned ParamCount
);
247 void si_llvm_dispose(struct si_shader_context
*ctx
);
249 void si_llvm_optimize_module(struct si_shader_context
*ctx
);
251 LLVMValueRef
si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context
*bld_base
,
256 LLVMValueRef
si_llvm_emit_fetch(struct lp_build_tgsi_context
*bld_base
,
257 const struct tgsi_full_src_register
*reg
,
258 enum tgsi_opcode_type type
,
261 LLVMValueRef
si_nir_load_input_tes(struct ac_shader_abi
*abi
,
263 LLVMValueRef vertex_index
,
264 LLVMValueRef param_index
,
265 unsigned const_index
,
267 unsigned driver_location
,
269 unsigned num_components
,
274 LLVMValueRef
si_llvm_load_input_gs(struct ac_shader_abi
*abi
,
275 unsigned input_index
,
276 unsigned vtx_offset_param
,
280 LLVMValueRef
si_nir_lookup_interp_param(struct ac_shader_abi
*abi
,
281 enum glsl_interp_mode interp
,
284 void si_llvm_emit_store(struct lp_build_tgsi_context
*bld_base
,
285 const struct tgsi_full_instruction
*inst
,
286 const struct tgsi_opcode_info
*info
,
288 LLVMValueRef dst
[4]);
290 /* Combine these with & instead of |. */
291 #define NOOP_WAITCNT 0xf7f
292 #define LGKM_CNT 0x07f
295 LLVMValueRef
si_get_indirect_index(struct si_shader_context
*ctx
,
296 const struct tgsi_ind_register
*ind
,
297 unsigned addr_mul
, int rel_index
);
298 LLVMValueRef
si_get_bounded_indirect_index(struct si_shader_context
*ctx
,
299 const struct tgsi_ind_register
*ind
,
300 int rel_index
, unsigned num
);
302 void si_shader_context_init_alu(struct lp_build_tgsi_context
*bld_base
);
303 void si_shader_context_init_mem(struct si_shader_context
*ctx
);
305 LLVMValueRef
si_load_sampler_desc(struct si_shader_context
*ctx
,
306 LLVMValueRef list
, LLVMValueRef index
,
307 enum ac_descriptor_type type
);
308 LLVMValueRef
si_load_image_desc(struct si_shader_context
*ctx
,
309 LLVMValueRef list
, LLVMValueRef index
,
310 enum ac_descriptor_type desc_type
, bool dcc_off
);
312 void si_load_system_value(struct si_shader_context
*ctx
,
314 const struct tgsi_full_declaration
*decl
);
315 void si_declare_compute_memory(struct si_shader_context
*ctx
);
316 void si_tgsi_declare_compute_memory(struct si_shader_context
*ctx
,
317 const struct tgsi_full_declaration
*decl
);
319 void si_llvm_load_input_vs(
320 struct si_shader_context
*ctx
,
321 unsigned input_index
,
322 LLVMValueRef out
[4]);
323 void si_llvm_load_input_fs(
324 struct si_shader_context
*ctx
,
325 unsigned input_index
,
326 LLVMValueRef out
[4]);
328 bool si_nir_build_llvm(struct si_shader_context
*ctx
, struct nir_shader
*nir
);